Course Handout - The History of the Psycholinguistic Flow Model

Copyright Notice: This material was written and published in Wales by Derek J. Smith (Chartered Engineer). It forms part of a multifile e-learning resource, and subject only to acknowledging Derek J. Smith's rights under international copyright law to be identified as author may be freely downloaded and printed off in single complete copies solely for the purposes of private study and/or review. Commercial exploitation rights are reserved. The remote hyperlinks have been selected for the academic appropriacy of their contents; they were free of offensive and litigious content when selected, and will be periodically checked to have remained so. Copyright 2010, High Tower Consultants Limited.


First published online 07:26 BST 3rd May 2002, Copyright Derek J. Smith (Chartered Engineer). This version [HT.11 - reinstate lost links] dated 12:00 8th December 2010

An earlier version of this material appeared in Smith (1997; Chapter 5).


1 - Psycholinguistic Modelling in Historical Context

The first sustained use of modelling techniques to help explain human communication was by the nineteenth century aphasiologists, and the efforts of Wernicke (1874), Kussmaul (1878), Lichtheim (1885), and Freud (1891) provide particularly good examples of what could be achieved in this way. However, in reviewing the achievements of these so-called "diagram makers", Head (1926) was largely dismissive, and human communication remained out of favour as a study area for modelling until the invention of the computer prompted a number of first generation models of attention and memory in the 1950s and 1960s. Here are some of the early efforts:

To see Craik's (1945) hierarchical model of biological control, click here.

To see Broadbent's (1958) "filter" model of attentional processing, click here.

To see Atkinson and Shiffrin's (1971) model of memory processing, click here.

Another influential early worker was the young British psychologist John Morton. He was studying the perceptual factors affecting word recognition and reading, and summarised the processes he suspected of being at work in a processing model based around a mental dictionary. Here is how he put it:

"It seems reasonable to assume that when a particular word is available as a response there is an event in the nervous system in a particular place regardless of the circumstances leading to the word availability. Such a part of the nervous system can be called a 'neural unit' [and the] collection of units makes up a 'dictionary'." (Morton, 1964:217.)

It is this use of the word "dictionary" which is significant, because the dictionary concept is a natural metaphor for the structure of long-term semantic memory [Memory Glossary]. Words can readily be seen as "units" in a mental lexicon (i.e. word store) composed of many such units, and to use a word from this mental dictionary, you simply have to "look it up" somehow, just as you would with a real dictionary. This means activating that particular word unit beyond some sort of activation threshold, whilst at the same time ensuring that no other word unit is allowed to approach its own threshold. Morton's 1964 model proposes a single mental dictionary, activated by one or other of two main input routes, but also strongly influenced by context effects and conscious selection. To see the full entry for Morton (1964), click here.

Two other important early workers were Marshall and Newcombe (1966, 1973), who had been studying the clinical phenomenon of acquired dyslexia, that is to say, dyslexia arising from brain damage or disease in previously non-dyslexic adults. By painstakingly assessing what acquired dyslexics could and could not do, these authors concluded that they could be divided into two main groups. In the first group - the "surface dyslexics" - a word and its errors were predominantly visually related (thus "insect" might be misread as "insist"). However, in the second group - the "deep dyslexics" - such errors were predominantly semantically related (thus "speak" might be misread as "talk", or "sick" as "ill"). Marshall and Newcombe (1973) modelled these findings using block diagram notation, and explained their observations by proposing that multiple processing routes operated during reading, each responsible for a different aspect of the process - form, sound, meaning, etc - and each operating "in parallel" (that is to say, simultaneously). To see the full entry for Marshall and Newcombe (1973), click here.

Reading processes were also being modelled at about the same time using other techniques borrowed from computing. Gough (1972), for example, used a systems analysis tool known as "suite flowcharting" with great success to describe how cognitive resources were progressively deployed. To see the full entry for Gough (1972), click here.

Morton continued to develop this type of model, and in 1979 he allocated the name logogen (from the Greek words logos = "word" and genesis = "birth") to the processes which called forth whole words in response to stimuli. A logogen is not a word, note, but rather "the device which makes a word available" (Morton, 1979, p112). In the jargon of the dataflow diagram, it is part-process, part-datastore (or, more accurately, it is a process which contains and manages a datastore). To see the full entry for Morton (1979), click here.

2 - Modern Psycholinguistic Modelling

The Marshall-Newcombe-Morton diagrams were valuable in their own right, but typically showed only the main processing modules (six for Marshall and Newcombe, 1973; five for Morton, 1979). In the event, however, they regularly proved to be too small to explain all the available data, and by the early 1980s Morton and a neuropsychologist named Andrew W. Ellis had both produced 21-box "supermodels", and it is at about this level of complexity that things have since stabilised.

To see the full entry for Morton (1981), click here.

To see the full entry for Ellis (1982), click here.

These large models soon became known as "transcoding models", because they constantly forced theorists to consider (a) what sort of information might be flowing along the individual flowlines, and (b) how it was (or was not) being transformed by successive processing. One early use of this term was by Weigl (1974), and another was by McCarthy and Warrington (1984). For a full glossary definition, click here.

The best known transcoding model within mainstream psychology is that by Ellis and Young (1988), and the best known within speech and language therapy are those by Kay, Lesser, and Coltheart (1992) and Stackhouse and Wells (1997). However, we also strongly recommend the smaller Coltheart, Curtis, Atkins, and Haller (1993) for the sophisticated consideration it gives to the flow processes linking the processing modules, for this is an important aspect of cognitive modelling which usually gets overlooked.

To see the full entry for Ellis and Young (1988), click here.

To see the full entry for Kay, Lesser, and Coltheart (1992), click here.

To see the full entry for Coltheart, Curtis, Atkins, and Haller (1993), click here.

To see the full entry for Stackhouse and Wells (1997), click here.

3 - Theoretical and Clinical Status of Transcoding Models

The problem with the larger psycholinguistic models, of course, is that at first sight they can appear rather daunting. Nevertheless, they are all basically dataflow diagrams (DFDs) [e-tutorial], consisting of subsystems, processes, and information flows, and their basic purpose is to display the totality of cognition in as few words as possible. They are merely pictorial expressions of arguments which could just as well have been expressed verbally, and in the final analysis all they can do is convey their authors' current understanding of the phenomena being modelled. Moreover, the DFD format itself is recognised as having certain inherent limitations, including the fact that "we cannot possibly represent everything we want to do in the same format" (Morton, 1981 p388). They are also bad at presenting the order of events, that is to say, at showing which processes are operating simultaneously and which serially, and when and why. Indeed, Morton has always been suitably reserved about the value of transcoding models, having described his 1979 model thus:

"This model makes it easier to relate together a large number of experimental findings and so may be regarded as a useful expository device. I do not believe it is 'true' in any interesting sense of the word and it is certainly not unique." (Morton, 1979, p109; bold emphasis added.)

The average transcoding model is "up for debate", in other words, and students are encouraged to evaluate and criticise them as vigorously as they would any other form of argument.

Speech and language pathologists are not wholly convinced of the utility of the available psycholinguistic models, either. Most hold that cognitive models are at best only partly useful in the clinic. Lesser (1987), for example, admits that by analysing "the mental processes underlying language into dissociable components or modules" (p189) transcoding models force you to address the basic organisation of neural processing, but she then argues that they have little to say on the higher matters of communication, such as in the realms of discourse [glossary] and pragmatics [glossary]. Similarly, Bryan (1995) describes the notion of modularity as "the underlying supposition" of cognitive neuropsychology, but complains that "clinicians are now asking not just what are the errors but why did they occur and 'where' might the difficulties in processing be" (p14; emphasis original), and Hillis (1993) puts it this way: ".....:there is nothing within the models of normal cognitive processes that would alone support the introduction of specific intervention strategies" (p6; italics original). What clinicians really need, she writes, is a theory of intervention, not of processing.

Cognitive neuropsychology is still actively refining its models in the hope of countering at least some of these criticisms, but is hampered in its efforts (a) by the sheer impenetrability of the underlying principles and mechanisms of cognition, and (b) by the fact that every clinical case is unique. Symptoms tend to be highly individual to a particular patient, rarely cleanly localisable, and often downright contradictory. Issues of current debate are whether the models should allow for separate processes for name, face, and picture identification, and even whether there is one semantic system or several. Only time will tell how much further they can be developed, and how much closer they can get to becoming Hillis's theory of intervention.

4 - References

See the Master References List